Mini toolsWan Streamer v0.2
By Alibaba
Model family: Wan
Wan-Streamer v0.2 is a native-streaming, end-to-end model that unifies text, audio, and video understanding and generation within a single autoregressive Transformer for real-time duplex conversation. It uses a Thinker-Performer architecture: a low-latency Thinker handles streaming perception, language and state updates, and decoding, while a multi-GPU Ulysses-style context-parallel Performer handles the heavier video latent generation. This split lets v0.2 raise output resolution from 192x336 to 640x368 at 25 FPS while preserving about 200ms model-side latency and roughly 550ms total interaction latency under a typical network budget. Not tied to a fixed avatar, it can embody any character describable in natural language, including humans, pets, and anime figures, supporting both close-up talking-head calls and scene-grounded mid-shot conversations where posture, gaze, hands, and surroundings stay legible.
New
Multimodal
Overview
Wan-Streamer v0.2 is an end-to-end Transformer model for real-time duplex audio-visual conversation. It listens, watches, understands, and responds with synchronized speech and video at roughly 550ms total latency, about 200ms model-side. Not tied to a fixed avatar, it can embody any character described in natural language, from humans to pets to anime figures. Version 0.2 raises output to 640x368 at 25 FPS, up from 192x336, enabling scene-grounded mid-shot interactions alongside close-up calls.
About Alibaba
Chinese e-commerce and cloud leader behind Taobao, Tmall, and Alipay.
View Company Profile
Last updated: July 10, 2026
